Skip to content

Alibaba opens a 14-day preview of Qwen4-Max with a 1-million-token context window

· by

The short version

Alibaba's Qwen team opened a time-boxed preview of Qwen4-Max on November 11, 2026, offering a 1-million-token context window through its Token Plan subscription, with no benchmarks, parameter count, or open weights disclosed.

Alibaba opens a 14-day preview of Qwen4-Max with a 1-million-token context window

Alibaba's Qwen team is letting developers try its next flagship model for exactly two weeks, then pulling it, with a context window of 1 million tokens, per Alibaba Cloud's Model Studio documentation, and no benchmark numbers attached to back up the claim.

What

Billed as qwen4-max-preview, the model became available at 10:00 on November 11, 2026, and is scheduled to retire at 24:00 on November 25, 2026, both in UTC+8, according to Alibaba's documentation. It accepts text, image, and video input for jobs like screenshot analysis, document understanding, and video analysis. A maximum output of 393,216 tokens is shared between the model's internal reasoning and its final answer, on top of that 1-million-token context window. Coding, office productivity, frontend design, and what Alibaba calls complex agentic tasks are the stated target uses, alongside support for web search, tool calling, structured outputs, and an adjustable reasoning_effort parameter. Developers can call the API using either the OpenAI or Anthropic SDK format, per the documentation. None of that is reachable through a metered, pay-as-you-go key during the preview window; a Token Plan subscription, sold in Personal or Team tiers, is the only path in.

No published numbers to check the claim against

Most model launches invite a side-by-side against GPT-5 or Claude's current lineup. This one does not: Alibaba's preview page carries no benchmark table, no stated parameter count, no license terms, and no downloadable weights. That gap matters for a company whose prior Qwen releases have often shipped open weights alongside the announcement. Whoever is sizing up Qwen4-Max for a production workflow is working from Alibaba's own feature list alone, not a third-party or self-reported score, and cannot even run a quick API trial, since the subscription gate rules out testing through a single metered call the way a new OpenAI or Anthropic model normally allows.

Context and reactions

Alibaba's Qwen line has moved fast through 2026, and several of those releases shipped open weights that outside developers could download and run themselves. Two choices break that pattern here: the preview disappears in two weeks by design, and it sits behind a paid subscription tier instead of a standard API product. Alibaba has not explained either decision on the documentation page, and no outside coverage of the preview had surfaced benchmark data as of this writing.

What to watch next

Whether Alibaba publishes a benchmark table or opens pay-as-you-go API access before the preview retires on November 25, and whether open weights follow once, or if, the model moves out of preview, matching the pattern of several earlier Qwen releases.

Sources